Hajra Naz
Can any company, small or large, topple Nvidia’s dominance in AI chips? Maybe not. The gap is wide, and the ecosystem around Nvidia is strong. Even then, there is a fortune in this market. Andy Jassy said so this week. He explained that hundreds of billions in revenue remain open for companies that grab even a small slice. The AI compute market keeps growing at record speed. High-performance GPU demand keeps climbing with every new AI model. This is why Amazon keeps pushing deeper into this space.
The company presented its next version of the Trainium AI chip at AWS re:Invent. The new chip is called Trainium3. It offers four times the speed of Trainium2. It uses less power. It improves energy efficiency for heavy training workloads. It positions Amazon for bigger deals in cloud AI. It also sends a clear message to the AI hardware market.
Jassy shared more on X. His comments showed why Amazon feels confident about this move.
Trainium2’s Current Momentum
Jassy said Trainium2 already has strong momentum. He said the business has a multibillion-dollar revenue run rate. He said more than one million Trainium2 chips are in production. He said more than one hundred thousand customers use Trainium2 for Bedrock work.
Bedrock is Amazon’s AI development platform. It lets companies pick and run many AI models. It powers enterprise AI tools. It supports generative AI pipelines. It makes adoption easier for companies dealing with new AI workloads.
Jassy said customers choose Amazon’s AI chip for a simple reason. He said Trainium2 offers a better price-to-performance ratio than other GPU choices in the market. He believes it delivers strong speed. He believes it costs less. He believes cloud customers want that mix. Amazon has used this playbook before. The company builds its own hardware. It offers lower pricing. It attracts a massive scale.
Read More: Amazon raises pressure on competitors through on-site Nvidia AI Factory systems
Anthropic’s Huge Demand for Trainium2
Matt Garman, CEO of AWS, added more detail in an interview with CRN. He explained that one customer fuels a large section of those billions. That customer is Anthropic.
Garman said AWS sees huge traction from Trainium2 inside Anthropic’s work. He pointed to Project Rainier. This project includes more than five hundred thousand Trainium2 chips. Anthropic uses them to train its next generation of Claude models. These models need huge compute. They need stable performance. They need lower costs at scale. Trainium2 supports these goals.
Project Rainier spreads across multiple U.S. data centers. It serves Anthropic’s rising demand for AI training power. It went live in October. Amazon invests in Anthropic. In return, Anthropic picked AWS as its main training partner. Anthropic also shows up on Microsoft’s cloud. That part uses Nvidia GPUs.
OpenAI also uses AWS now. That partnership is new. It brings more training and inference work to Amazon’s cloud. But this partnership did not add much revenue to Trainium. AWS said this part runs on Nvidia hardware.
Read More: Trump Says ‘No One Gets Nvidia’s Most Advanced Chips,’ But Huang Warns China Will Win The AI Race
Why Nvidia Still Holds the Strongest Lead
Only a few U.S. tech giants have enough pieces to build a serious challenger to Nvidia. Google, Microsoft, Amazon, and Meta fall into that group. They hold chip design teams. They hold custom networking engineers. They hold advanced interconnect tech.
Nvidia also holds key advantages. The company bought Mellanox in 2019. That buyout gave Nvidia deep control over high-speed networking tech. That control matters for AI clusters and high-bandwidth compute.
Nvidia also owns CUDA. CUDA is its proprietary software system. CUDA lets AI apps use Nvidia GPUs for parallel computing. CUDA also powers model training tools and advanced kernels. Rewriting an AI application for a non-CUDA chip takes a huge effort. Many companies prefer sticking with Nvidia because switching disrupts workflows. This keeps Nvidia ahead in the silicon race.
Read More: Nvidia’s message to Google: Nice try, we’re still on top
Amazon’s Strategy for the Future
Amazon may have a path that avoids this problem. The next Trainium chip, called Trainium4, will work inside systems that also run Nvidia GPUs. Trainium4 will sit beside Nvidia hardware. This setup could draw customers who want flexibility. It could reduce friction around CUDA. It could help companies mix hardware without heavy rewrites. It might also keep Nvidia dominant, but only inside AWS infrastructure. Time will reveal the impact.
None of this may bother Amazon. The company already expects multibillion-dollar revenue from Trainium2. Trainium3 pushes performance higher. Trainium4 moves toward hybrid TPU-GPU style systems. The trend points toward stronger AI infrastructure on AWS. That may be a win on its own. Amazon gains more cloud computing income. It gains stronger partnerships. It gains AI leadership inside enterprise cloud markets.
For Amazon, that may be enough. The chip does not need to beat Nvidia. It only needs to grow. And it is growing fast.
